16 Oct The biggest AI breakthroughs over the past 10 years

Artificial Intelligence has always been an object of admiration and damnation. Skim through AI’ summers and winters, and you’ll understand that we’ve always taken sides. On the one hand, we’ve been amazed by what man has created – “thinking machines”, “machines that could behave intelligently”, “machines with the same level of intelligence as man.” On the other hand, we’ve always feared man’s own creation. It happened with every Industrial Revolution. So as Artificial Intelligence keeps ascending at full speed, it’s only natural for us to grow  even more anxious.

The ascent of Artificial Intelligence is leaving us partly hungry for more automation, partly afraid of too much computerization. And that’s where the paradox lies: we dismiss technologies that ask us to put in effort; we want them to be friendly to us; we feel good when they do most of our work; we relish the perks of having intelligent systems in our homes, cars and jobs, yet we fear mass unemployment and the frightening scenario where AI takes over the world and gets out of our control.

This article does not focus on clarifying whether fearing the AI supremacy is legitimate or not (for that, I encourage you to listen to computer scientist Kai-Fu Lee beautifully scrutinizing the matter in this TED Talk). 

Instead, let’s revisit AI’s main breakthroughs in the past 10 years, just to remind us that, in the right hands, technology can truly make our lives better.

1997 | Deep Blue

IBM’s Deep Blue was the first computer in the world to win against a world champion. In a pair of six-game chess matches that lasted several days, Deep Blue beat Garry Kasparov proving that computers can indeed handle complex calculations. Just so you know, Deep Blue could calculate 200 million possible chess positions per second.

For computer scientists, this was a promising start to discovering new medical drugs, creating the broad financial modeling required to identify trends and do risk analysis, handling large database searches.

In 2008, a poker-playing AI program called Polaris beat two champs during the Man-Machine Poker Competition in Las Vegas. Later on in 2011, IBM’s Watson beat Jeopardy! champions Brad Rutter and Ken Jennings, winning the $ 1 million grand prize.

1998 | Furby

Furby was the first domestically-aimed robot, a sort of primitive AI designed by Tiger Electronics. The soon-to-be one of the most popular toys in the world introduced the concept of humans chatting with robots. Furby entered people’s homes and became part of their daily lives, selling in 40 million copies over 3 years.

Furby could chat with kids, answer their questions and learn English. When taken out of the box, Furbies would talk their own garbled language called Furbish, but then they would adopt English words into their vocabulary. They were programmed to mimic the process of language development and learn from conversations with humans.  

2002 | Roomba

The original iRobot Roomba vacuum cleaner came out in 2002, when AI was not that advanced. iRobot CEO Colin Angle had been studying and building robots as an MIT student, trying to make machines intelligent. After MIT, Angle worked with the US Department of Defense to build robots that would clear minefields. This collaboration resulted in an AI system that enabled robots to check every area of a field and ensure coverage. The solution for cleaning and coverage at a low cost inspired Angle to build an intelligent vacuum for consumers.

Roomba could vacuum the house when you weren’t home. With the help of sensors, it could move around the house by itself. Whenever it sensed an obstacle, it was able to change directions. At that time, machine learning was at an early stage, so the decision-making process of this device was quite basic. Whenever it bumped into a wall, for instance, it would try various tactics, like using a different angle or route again, without putting cleaning on hold. It could detect dirty spots and avoid falling down the stairs. Years later, the Roomba could systematically clean the floor using onboard mapping and navigation software.

2010 | Siri

Siri marked the beginning of the virtual assistants era. Two months following its release as a mobile application by Nuance Communications, Siri was acquired by Apple and then integrated in the iPhone 4S.

Siri was the first intelligent personal assistant to replace keyboards and touch screens. In a highly natural voice, Siri would make recommendations, schedule events, answer questions, adjust device settings and run searches. The more users would interact with Siri, the more it would adapt to their individual preferences.

It took Amazon 4 years to come up with a similar voice interface (Alexa), whereas Microsoft’s Cortana and Google Assistant caught up later on, with less effective results.

2015 | ImageNet Challenge

In 2015, Microsoft and Google machines beat humans at image recognition at the sixth edition of the  ImageNet Large Scale Visual Recognition Challenge. The machines one-upped humans due to deep learning algorithms that enabled them to identify images and objects in over 1,000 categories. The algorithms were derived from various versions of artificial neural networks which mimicked the way the human brain works.

This exciting new breakthrough allowed intelligent systems to automate tasks that require recognition of an object or person and then make a decision about how to proceed based on that recognition.

2016 | AlphaGoZero

The extraordinary success of Deep Mind’s AlphaGoZero at the game of Go advanced the practical possibilities of AI to an unprecedented level. AlphaGoZero is a machine-learning system that in a short amount of time became a world master of the complex game of Go.

This was the first AI to learn on its own. More precisely, AlphaGoZero played completely random games against itself and learned from the results. Thus, it was able to beat world champion Lee Se-dol by 100 games to 0. At that time, DeepMind lead researcher David Silver said that achieving this level of performance in the complex context of Go “should mean that we can now start to tackle some of the most challenging problems for humanity.“

2018 | OpenAI

Elon Musk’s research group OpenAI created a bot that beat top players in Dota 2.  The bot was actually an algorithmic team, called OpenAI Five. Each algorithm used a neural network to learn how to play the Dota 2 game and cooperated with its AI teammates. The algorithms learned by playing against different versions of themselves and did not use imitation learning or tree search.

OpenAI’s breakthrough started a major new direction for AI since algorithms usually operate independently. According to the researchers at OpenAI, “this is a step towards building AI systems which accomplish well-defined goals in messy, complicated situations involving real humans.”

Thanks to these breakthroughs, many AI milestones have now been reached, making us take seriously the possibility of witnessing superintelligence in our lifetime. With AI creating artwork, writing literature, deciphering Vatican’s Secret Archives, predicting earthquakes or anticipating patients coming out of the coma, superintelligence seems more like a question of “when” than “if”.